An evaluation of those models which were introduced in part I of this report was carried out. Based on ECMWF model cloud and precipitation profiles a relative comparison of doubling-adding and Eddington models revealed that apart from situations with excessive amounts of snow both models agree within radiometer noise level, i.e., within 0.5 K. The treatment of sub-gridscale cloud variability was identified as an important source of error. At frequencies above 37 GHz, a separation of the average cloud profile into three sub-profiles, namely cloudfree, ice and water portions with corresponding maximum fractional coverage, seemed favorable. An intercomparison with radiative transfer codes from independent research groups also revealed the good performance of the two models. In conclusion, the Eddington model showed acceptable accuracy with a 30 times higher computational efficiency thus being well suited for global application < 100 GHz.
}, year = {2004}, month = {2004}, language = {eng}, }